Computer Simulation

Module 6: Generating Uniform

Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Introduction

M6-001
 Module Overview
Last Module: We covered basic
(and advanced) Arena concepts.

This Module: We’ll do a deep dive

to investigate how to generate
Unif(0,1) pseudo-random numbers.

Idea: It’s a deterministic cookbook!

To Serve Man:
www.youtube.com/watch?v=dk01eeKMD_I

M6-002
 Module Overview
1. Introduction • This lesson
2. Some Lousy Generators
3. Linear Congruential
Generators
4. Tausworthe Generators
5. Generalizations of LCGs
6. Choosing a Good Generator
— Some Theory
7. Choosing a Good Generator
— Statistical Tests
M6-003
 Introduction
niform(O 1 random number are the ke to random ariate
generation in imulation - ou tran form uniform to et other RV .

Goal: Gi e an algorithm that produce a equence of p udo-randon

nun b ~ (PRN R-1 R2 ... that appear to be i.i.d. Unif O 1 .

De ired propertie of algorithm

■ output appear to be i.i.d. Unif O 1
■ ery fa t
■ abilit to reproduc an equence it o nerate
M6-004
 Intro (cont’d)
Cla e of Unif O 1 Generator
■ Some lou enerator
■ output f rand m de i e
■ table of random number
■ mid quare
■ Fibon 1
■ Linear congruential mo t commonl u ed in practice
■ Tau worthe linear recur ion mod 2
■ H brid
M6-005
 Summary
This Time: Discussed what’s
coming up in this module on
Unif(0,1) random number
generation.

Next Time: We’ll actually spend a

little time looking at poor
generators!

M6-006
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Some Lousy Generators

M6-007
 Lesson Overview
Last Lesson: We introduced the
topic of Unif(0,1) generation. It’s
the key to all knowledge!

This Lesson: We’ll spend some

time talking about poor generators.

Idea: What could possibly go

wrong? •

M6-008
 Some Lousy Generators
a. Random Device
ice randomne propertie . Howe er Unif O 1 equence torage
difficult o it tough to repeat exp riment.

Example ·
■ flip a oin
■ particle count b Geiger counter
■ lea t ignificant diait of atomic clock

M6-009
 More Lousy Generators
b. Random umber Table
Li t of digit upplied in table .

■ A Milli n Rando,n Di it ith 100,000 rn al D i t

www.rand .org/conten dam/ rand/pubs/monograph_reports/ MR1418/MR1418.digits.pdf

Cumber ome low table too mall - not er u eful.

On e tabled no lonoer random.

M6-010
 Even More!
c. id-Square ethod J. on eu1nann

Idea: Take the middle part of the quar of the pre iou random
number. on eumann wa a fun-lo in gu but method i terrible!
Example: Take R - == ·/ 10000 v· where are integer < 10000.
t d o == 632· then 6 22 ➔ 98342 ·
So 1 == 3 · then 9 2 ➔ 9 7075 ·
So 2 == 707 etc ...
U nfortunatel po 1t1 e erial orrelation in Ri
Al o oc a ionall degenerate · e.g. con ider
M6-011
· == 0003.
 Still More Bad Boys
d. Fibonacci and Additi e Congruential Generator

1, ·- 1 + i- 2 dm . == 1 2 ....
)

where R· == ·/ i the ,nodulu -1 o are d and

== bm d iff a i the remainder of b/m e.g. == 1 m d 7.
Problem: Small number follow mall number .

Al o it not po ible to oet ·_ 1 < ·+ 1 < · or

M6-012
i < ·+ 1 < i -1 which hould o cur w.p. 1/3 .
 Summary
This Time: Talked about some Unif(0,1)
rotten apples, as well as some reasons
for the worms.

Next Time: Finally, a good generator!

The linear congruential generator.

M6-013
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Linear Congruential
Generators
M6-014
 Lesson Overview
Last Time: Discussed some awful
Unif(0,1) generators.

This Time: We’ll look at a really

good one – the linear congruential
generator.

Variations of the LCG are the

most commonly used generators
in practice.

M6-015
 LCGs
LCG are the 1no t widel u ed generator . The e are pretty good
when implemented properly.

1, - ·- 1 + mod where Oi the eed.

R · == ·/ '
. == 1. 2
J

Choo e arefull to get oood tati tical quality and lono

p riod or l l n tJ i.e. time until LCG tart to repeat it elf.

If == 0 LCG i called a n ultipli ati oenerator.

M6-016
 Trivial Example: For purpo e of illu tration con ider the LCG

i - ( ·-1 +3 m
If 0 == 0 we ha e 1 == ( o+ 1n d -- · continuing

0 1 2 3 4 s 6 7 9
0 3 2 s 4 7 6 1 0 3
3 2 4 7 6 1 3
R· 0 - - - - - - 0 -

o that the equence tait repeating with == 0.

Thi i afull-p riod n rator ince it ha c cle length m == .

Generali peaking full-period i a good thing. □
M6-017
Plot of (Xi-1 , Xi) ~

-
~

(4,7)
j

?_
,,,,
/ (7,6) /
(2,5),..,, / / ~
/ (5,4) I/
(0,3) I / / ~
I
/ (3,2) ,L_
,,,,
(-2,1) / / (6,1)

,====,
/ ~ (1,0) V ·~,I
....
~

-,e·

/ "'
The random
numbers fall
mainly on the
planes

M6-018
 Easy Exercise
Consider the generator

Xi = (5 Xi -1 + 2) mod 8

(which is very similar to the previous

generator).

Does this generator achieve full

cycle? Explain.

Now that’s odd. No it’s not! •

M6-019
 Better Example de ert i land venerator : Here our old 16 07
implementation BFS 19 7 whi h I e tran lated from FORTRA .
It work fine i fa t and i full-period with c cle length > 2 billion

i == 1 07 i -1 m 231 - 1 .

Algorithm: Let O be an integer eed between 1 and 231 - 1.

For · == 1 2

} +-- l ·-1 / 12777 J integer di i ion lea e no remainder

i +-- 1 07 i -1 - 12777 } ) - 2 6K
if · < 0 then et · +-- i + 21 7 3 7
R · +-- i * . 12 7 E-10

M6-020
 Example: If O == 12 7 then K == and
1 +- 1 07 [12 7 - 12777 ] - 2 == 1 00
o that R1 == 0. 21 3 . D

Sta tuned for ariou wa to a e the quality of P enerator .

M6-021
 Fir t of all what can go wron with LCG .
■ Somethin like · == ( ·- 1 + 2 m d i not full-period
ince it onl produce e en integer .
■ Something like · == ( i -1 + 1) m d i full-period but it
produce ery non-random output: 1 == 1 2 == 2
'
etc.
■ In an a if m i mall ou 11 get quick c clino whether or not
the oenerator i full period. Small ould mean an thing le
than 2 billion or o !
■ And ju t becau e i big ou till ha e to be careful. Some
ubtle problem an ari e. Take a look at RA U....

M6-022
 Example: The infa1nou RA D oenerator
·- 1

wa popular durino the 1960 .

Here what R i -2 R i -1 R-i look like if ou plot them in 3-D tol n
from Wikipedia . If they were t1ul iid Unif O 1 ou d ee dot
random! di p r ed in the unit cube. But in tead the random number
fall entire! on 15 h perplane not good .

..
."
.
·"
.,
"
"0

M6-023
 Exercises:
D Implement RA DU and ee how it doe . That i plot R ·-1
R,i for · == 1 2, ... 1000 and ee what happen . Tr a few
different . eed and ma be ou 11 ee ome h perplane . You 11
e11ainl ee them if ou plot (Ri-2 , R ·- 1 R · in 3-D.
FJ ow do the a1ne thing with the 16 07 enerator. You probabl
won t be able to ee an h perplane here.

M6-024
 Summary
This Time: Discussed the linear
congruential generator.

Next Time: We’ll talk about an

alternative good PRN generator – the
Tausworthe generator.

M6-025
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Tausworthe Generators

M6-026
 Lesson Overview
Last Time: Discussed LCG PRNs,
OMG! QED, BTW! LOL! •

This Time: We’ll look at a

“competing” good generator – the
Tausworthe generator.

This class of generators seems to

be more popular with computer
science folks.

M6-027
 Tausworthe Generator
Defin a equence of binar digit B 1 B , . . . b

where J == 0 or 1. Look a bit like a generalization of LCG .

U ual i1nplementation a e computational effort :

B· == Bi-r + Bi-q ) m d2 == Bi-r XOR Bi-q O< < q .

M6-028
 Obtain

To initialize the B i equence p cif B1 B2 . . . Bq.

Example Law 2015 : == 3 q == ; B1 == • • • == B == 1. Obtain

B i == (B·_3 B -_ 1nod 2 = Bi-3 XOR B i- >
BB == B3 XOR B1 ) == 0 B1 == B4 XOR B2 ) == 0 etc.
1111 1000 1101 1101 0100 0010 0101 1oq1 1111 □

The period of 0-1 bit i alwa 2q - 1 == 31.

M6-029
 How do we go from Bi to nif O 1

Ea wa : U e -bit in ba e 2 /2 and con ert to ba e 10.

Example: et == in pre iou example and get

1 1 1
1111 1000 11012 11012 , ... ➔ -1 l □
16 1

Lot of potential for Tau worthe generator . ice prop rtie

includino- long period fa t calculation.

M6-030
 Summary
This Time: Discussed Tausworthe
generators, which are alternatives to
LCGs.

Next Time: We’ll look at

generalizations of LCGs, some of
which have incredible properties.

M6-031
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Generalizations of LCGs

M6-032
 Lesson Overview
Last Time: We studied the
Tausworthe PRN generator, which
actually looked a bit like an LCG.

This Time: We’ll look at

generalizations of LCGs.

Some of these have incredible

properties.

M6-033
 A Simple Generalization:
i == ( I:]=l · i- J)m where the ai are con tant .
Extreme! large period po ible up to m q - 1 if parameter are
ho en properl . But watch out! - Fibonacci i a pecial ca e.

Combinations of Generators:
Can combine two generator 1, 2 ... nd 1, 2 ... to con truct
Z1 Z .... Some ugge tion :
■ Set Z i == i + · mod m
Danger Will Robinson!
■ Shufflino- Properties are difficult
■ Set Zi == · or Zi == i to prove!
M6-034
 A Really Good Combined Generator due to L Ecu er 1999
and di cu ed in Law 2015 .

Initialize 1o 11 12 2 o 21 2 . For · > et

1i 1 403 5 0 1 i- 2 - 10 72 1 ·_ 3 m (232 - 209

2i i -1 - 1 370 5 9 2 ·_3 m (232 - 22 53
'l 1i - 2 . 232 - 20 )
R-i ·/ 232 - 2 9)

A raz a thi generator look it actual! pretty imple work

well and ha an amazing c cle length of about 2191 !

M6-035
 Some Remarks
It is of interest to note that Matsumoto
and Nishimura have developed the
“Mersenne Twister” generator, which
has period of 219937 – 1 (yes, that's a
prime number) and works great!!

You’ll often need several billion

PRN’s, but never anything over 2100.

All standard packages use one of

these good generators.
M6-036
 Summary
This Time: Discussed generalizations
of certain PRN generators, and gave
a couple awesome examples.

Next Time: We’ll go over some

theoretical considerations when
deciding on a good generator to use.

M6-037
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Choosing a Good Generator

– Some Theory
M6-038
 Lesson Overview
Last Time: Gave a couple of
generators with extremely long
cycle times.

This Time: We’ll discuss some

PRN generator properties from a
theory point of view.

I’ll present an amalgamation of

typical results that I won’t really
hold you responsible for.
M6-039
 Here are ome 1ni cellaneou re ult due to Knuth and oth r that are
helpful in determinin 0 the quality of a P generator.

Theorem: The enerator ·_ 1 1. d 2n > . can ha e

c cle lenoth of at m t 2n- 2 . Thi i achie ed when o i odd and
+ . or == k + for 01ne k.
Example BC .
i == 1 i -1 m d 6

X X1 X2 X3 X4
1 1. 1 ~1 17 1
..... 1 4 ... 2
1 3
4 2 36 2 4 Really short periods! •
M6-040
 Lot of th e t_pe of c_cle length re ult .
Theorem: , , - . ·_1 + · d > 0 ha full c 1 if i
and m are relati el prime· ii - 1 i a multiple of e er prin1e
which di ide · and 111 - 1 i a multiple of 4 if 4 di ide

Corollary: · == ( i -l + )m d 2n > 1 ha full cy le if ·

i odd and + 1 for ome l .
Theorem: Then ultipli ati enerator ·-1 n with
prime ha full period - 1 if and only if i di ide m- l - 1 ·
and ii for all inte er · < - 1 doe not di ide , , - .
Remark: For , == 231 - 1 it can b hown that 5 4 600 000
multiplier ield full period the be t of whi h i == 950 706 7 6
Fi hman and Moore 19 6 .
M6-041
 Geometric Considerations

Theorem: The /·-tuple R ·i ... R· k-l · > 1 fro1n

1nultiplicati e generator lie on parallel h perplane in [O 1] .
Th followin g ometri quantitie ar f int r t.
■ Minimum numb r of h perplane in all direction . Find the
multiplier that 1naximiz thi numb r.
■ Maximum di tan e b t en parallel h perplane . Find the
1nultipli r th t mini1nize thi nun1ber.
■ Minimum Eu lid an di tan e b t en adjac nt k-tuple . Find
th 1nultipli r that maximize thi nu1nber.
Remark: The RA D oenerator i particular! bad ince it lie on
onl 15 hyperplane .
M6-042
 OtlCJn11
an al o look t one- tep erial correlation .
RN
Serial Co11· lation of LCG Greenber0 er 1961 :

(- )2) + - +6
Thi upper bound i ery mall for m in the ran e of 2 billion and
a == 1 7.

Lot of other theor con ideration that can be u ed to e aluate the

perfo1mance of a particular PR generator.

M6-043
 Summary
This Time: We saw a mishmash of
PRN generator theoretical properties
– just to give you a flavor (not to
hold you responsible, necessarily).

Next Time: We’ll go over some

statistical considerations when
choosing a good generator. These
may be a bit more intuitive and
relevant.

M6-044
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Choosing a Good Generator

– Statistical Tests, Intro
M6-045
 Lesson Overview
Last Time: Plowed thru some
miscellaneous theoretical
properties of PRNs.

This Time: We’ll give an overview

on statistical tests for goodness-
of-fit and independence.

The chi-squared test will really

give you fits!

M6-046
 Statistical Tests Intro
We’ll look at two classes of tests:
Goodness-of-fit tests — are the
PRNs approximately Unif(0,1)?
Independence tests — are the
PRNs approximately independent?
If a generator passes both types of
tests (in addition to others I won’t
tell you about), we’ll be happy to
use the PRNs it generates.

M6-047
 Intro (cont’d)
All tests are of the form H0 (our
null hypothesis) vs. H1 (the
alternative hypothesis).

We regard H0 as the status quo,

so we’ll only reject H0 if we have
“ample” evidence against it.
(Innocent until proven guilty.)

Usually, we really want to avoid

incorrect rejections of H0.
M6-048
 Intro (cont’d)
When we design the test, we set the
level of significance
α = P(Reject H0 | H0 true).
Typically, α = 0.05 or 0.1, and is the
probability of Type I error.

We can also specify the probability of

Type II error,
β = P(Accept H0 | H0 false),
but we won’t worry about that just
now.
M6-049
 Summary
This Time: Presented an introduction
to hypothesis testing in the context
of selecting a PRN generator.

Next Time: We’ll discuss the chi-

squared goodness-of-fit test to
check whether or not the PRNs are
actually uniform.

M6-050
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Choosing a Good Generator

– Goodness-of-Fit Tests
M6-051
 Lesson Overview
Last Time: Quick intro to
hypothesis testing. Interested in
testing PRNs for uniformity and
independence.

This Time: We’ll discuss the chi-

squared goodness-of-fit test to
check whether or not the PRNs
are actually uniform.

There are many g-o-f tests, but

this is the most tried-and-true.
M6-052
 2 Goodness-of-Fit Test

Di ide the unit inter al into k cell ubinterval . If ou boo e

equi-probablecell [O.! ) [! ~ j • • · [k,; 1 1] thenapa1ticular
ob er ation Rj will fall in a parti ular cell with prob 1/ k.
Tall how man of the ob er ation fall into the k ell . If
Oi # of Rj in cell · then ince the Rj are i.i.d. we can ea il
ee that O · rv Bin . !) .
== 1 2. . . . k.
Thu the expected number of Rj to fall in cell · will be
i [Oi] == / . == 1 2 ... , k.
M6-053
 2 Goodness-of-Fit Test

We reject the null h pothe i Ho if the Oi don t match the i well.

The / goodne -of-fit tati tic i

k 2
2
0 L--
i =l

A large alue of thi tati ti indicate a bad fit.

M6-054
 2 Goodness-of-Fit Test

In fa t we r '.it the null h pothe i Ho that the ob et ation are

uniform if /6
> /2 k - l where 12 k - l i the appropriate 1 -
quantile from a 2 table i.e. P ~- l < 2 k - l ) == 1 - '.

If 5< /; k-l wefuil tor '.i t Ho.

U ual recommendation from baby tat cla : For the I o-o-f te t to

work pick k , uch that i > and at lea t r,O. But. ..

M6-055
 Unlike what ou learned in bab tat la when we te t PR
oen rator we u uall ha a hu number of ob r ation at lea t
million with a large nu1nber of cell k. When k i large we can u e
th approximation

2
k- 1 ~ (7 - 1 1- 2 +
9 k- 1 " - 1
where i the appropriate tandard normal quantile.
Remarks: 1 16 07 PR enerator u uall pa e the -o-f te t ju t
fine. 2 We 11 how how to do g-o-f te t for other di tribution later
on - ju t doing uniform PR for now. 3 Other -o-f te t :
Kohnooorov- mimo te t Ander on-Darlin te t etc.
M6-056
 Illustrative Example: == 1000 b rvation k == int r al .

int r al [0 0.2] 0.2 0.4] 0.4 0. 6] 0.6 0. ] 0. 1.0]

200 200 _oo 200 200
0- 179 20 222 199 192

2 2
0 .27. k- 1 0.0 9.

Since 6
< 2 k-l we fail to reject Ho and owe 11 la u1ne that the
ob er ation are appro imatel uniform. □

M6-057
 Summary
This Time: Showed how to do a chi-
squared g-o-f test for uniformity of
PRNs. It’s fun! It’s nutritious!

Next Time: Independence Day!

M6-058
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Choosing a Good Generator

– Independence Tests, I
M6-059
 Lesson Overview
Last Time: Discussed the chi-
squared g-o-f test for uniformity of
PRNs. It gave us the fits!

This Time: We’ll look at so-called

“runs” tests for independence of
the PRNs. Czech it out!

(Next time, we’ll look at

autocorrelation tests for
independence.)
M6-060
 Independence – Runs Tests
ow on ider th h poth i Ho : R1 R2 ~... R ar independ nt.

Let look at Three Little Bear example of coin to ino:

A. H T H T H T H T H T. . . negati e cotTelation

B. H H H H H T T T T T. . . po iti e correlation

C. H H H T T H T T H T. . . ju t right

M6-061
 Definition: A r ,n i a erie of imilar ob er ation .

In A abo e the run ar : H T H T . . . . man run

In B the run are: HHHHH TTTTT ery few run

In C: HHH TT H TT . . . . medium number of run

A run t twill reject the null h path i of ind p ndence if there ar

too 1nan or too few run whate er that 1nean . There are ariou
t pe of run te t · we '11 di cu two of them.

M6-062
 Runs Test "Up and Down". Con ider the following PR .

.41 .6 . 9 . 4 .74 .91 .55 .7 1 .36 .30 .09. . .

If the uniform increa e put a +· if it decrea e put a - like H and
T . Get the equence

++ -- + - + --- ...
Here are the a ociated run :

++ -- + - + - - -
So do we ha e too man or two few run .
M6-063
 Let d note the total nu1nber of run up and down out of
ob ervation . in the abov example.

Amazino Fact: If i lar e a > 20 and the Rj are actual!

independent then
2 - 1 1 - 2 )
or ( 9 .

We 11 reje t Ho if i too bi or 1nall. The te t tati ti i

[ ]
Zo
✓
and we reject Ho if
M6-064
IZol > /2·
 Up and Down Example
Suppose that n = 100 and A = 55.

Then A is approximately

Nor(66.33, 17.46).

So Z0 = -2.71.

If α = 0.05, then zα/2 = 1.96, and

we reject H0 (i.e., reject
independence).
M6-065
 Runs Test "Above and Below the Mean ". Again con ider
.41 .6 . 9 . 4 .74 .91 .55 .71 .36 .30 .09...
If Ri > . put a · if Ri < O.. put a-. Get the equen e
- + + + + + + + - - - ...

Here are the a ociated run of v..1hi h there are B == 3:

+ + + + + + + ---

M6-066
 Fact: If i lar and the Rj are actuall independen~ then

B~
where 1 i th numb r of ob ervation > . and ·2 == - 1·

The te t tati tic i Zo == (B - E [B ] / J ·r B and we reject Ho if

IZo l > / 2·

M6-067
 Illustrat ive Example from BC uppo e that with the
followino +/ - equence.

- + + + + + + + --- + +-+-----
-- + + ---- + + -- + - + -- +-
Then 1 == 1 == 22 and B == 17. Thi implie that [B ] . 20. ·
and r B · . . And thi ield Z 0 == - 1.07.

Sine IZo l < I == l. 6 we fail tor ject th te t' owe an treat

the ob er ation a independent. □

Lot of other te t a ailable t r independence: 0th r run te t

orrelation te t bap te t poker te t birthda te t etc.
M6-068
 Summary
This Time: We ran off at the mouth
with runs tests for independence of
PRNs: “Up and Down” and “Above
and Below the Mean”.

Next Time: A bonus “autocorrelation”

test for independence!

M6-069
Computer Simulation
Module 6: Generating Uniform
Random Numbers

Dave Goldsman, Ph.D.

Professor
Stewart School of Industrial and Systems Engineering

Choosing a Good Generator

– Independence Tests, II
M6-070
 Lesson Overview
Last Time: We looked at a couple
of runs tests for independence of
PRNs

This Time: Autocorrelation tests

for independence.

This is sort of bonus material, but

we’ll be talking about autocorr
throughout the course, so let’s
take it out for spin here.
M6-071
 Correlation Test. A umino- that the R · are all Unif O 1 let
conduct a orr l Lti n te t for Ho: R· independent.

Wedefinethel, -1 orr latio, oftheR· b rr R· R-i +l

Ideall p hould equal zero. A oood e timator for p i gi en b

12 n -1
p"' - - ~ R-kR 1+ ..
-1~
k=l

p"' ~ or (o 1 7
-
1 )
- 1)2
under Ho .

The te t tati tic Zo == "' / ✓ and we reject if IZo I > /2·

M6-072
 Illustrative Example: Con ider the followin == 30 P
0.29 0.3 0.46 0.29 0.69 0.73 0. 0 0.74 0.99 0.7
0. 0.66 0.56 0.41 0.3 0.22 0.1 0.05 0.25 0.36
0.39 0.45 0.50 0.62 0.76 0. 1 0.97 0.72 0.11 0.5

After a li ttl algebra w get

- 19
""
p .9 0 and .r p ""
0. 1.

So Zo == o_gr.:o; Jo. 1 == 1. .
Sine IZo l < / == 1. w fail tor ject th t t meanin° that w
can treat the PR a independent. Of cour e == 0 i ort of
1nall and perhap thi deci ion will chanoe if we increa e . D
M6-073
 Summary
This Time: Autocorrelation tests for
independence of PRNs.

This completes Module 6 on PRNs.

Next Module: PRNs are too easy!

When we return, we shall be
promoted to random variate
generation!
https://getyarn.io/yarn-clip/49a98d2d-6852-4c74-8e06-5a87dac9e915

M6-074